Whats your settings?, remember that bigasp need some specific stuff, for example cfg at 3 and at the beginning of the prompt you must add "score_7_up"

sounds like you need the sdxl vae

@waterdrinker there model called'BigaspV2Lustify' exact same your merge percentage , but does it differ ?

you can make one yourself by merge.

@cptdan Thanks. Did a merge and works great

@Boxer766 did you just merge it with an inpainting model? or is there a base template model for inpainting

@Boxer766 do you have a link to your inpainting model? I'm interested

Does anyone have some tips on which models to merge for an inpainting version? I keep getting errors about incompatible models, size mismatches, etc.

Just use the DMD2 LoRA.

first step would be to decide what program you want to use:

- Forge (most beginner friendly)
- Automatic1111 (the standard for many)
- ComfyUI (superior control, not very beginner friendly)

you will find tons of good tutorials on how to set up each of the above on youtube! And if you are really lazy and don't know much about computers, use Pinokio.

Installation guide for beginners: https://stable-diffusion-art.com/install-windows/

Too many dislikes towards a very precise comments. The checkpoint is AMAZING for NSFW images, but it is almost impossible to generate full SFW stuff. Sometimes you want to generate teasing images, instead of full NSFW stuff, and a few adjustments to the checkpoint in order to achieve such thing would be amazing.

if you use dmd2 and have a cfg of 1 negative prompts are ignored. you can try increment cfg to 1.1 or 1.2 and hope that it takes your negative prompt into accountwithout burning your image. you can edit the config files of forge and automatic1111 to get more control over how fine you can adjust the sliders.

このモデルは素晴らしいです！
しかし、誰かインペインティングでの使い方を教えてくれませんか？
A111のインペイントタブで「fill」や「original」などを試してみましたが、どれを使ってもインペイントがひどくてぼやけてしまいます。
どなたか助けていただけませんか？

@ghost958 you need to make your own inpainting model by merging.

@cptdan Thank you. are there any you would recommend merging with. And what weight for A and B, if you know? Thanks so much

@ghost958 the procedure is the same from sd 1.5, just different models.

What are your preferred settings? Also do you tag more or nlp more?

Are you planning on sharing that model? I would love to try it out!

@tetonasenjoyer I might look into uploading it on Civitai, thanks for the suggestion. In the meantime, here's my workflow .json you can load into Comfy to make the same 70/30 merge: https://pastebin.com/Dh0Qdg00

The new LustiMix is also very good and seems to use a similar recipe: https://civitai.com/models/1352908/lustimix

@focalbluebell I'll try it out! Thanks for the info.

all I'm getting out of this model are blank images. looks like I wasted my time downloading this

Do you have ComfyUI? You can drag one of OP's images into it and all the nodes show up.

Download sdxl.vae.safetensors to the VAE folder.

Use high score numbers in the positive prompt (eg. score__7_up) and low score values in the negative prompt (eg. score_1, score_2 etc). Use further negative text such as 'deformed, mutated, logo' etc.

DPM++ 2M SDE for sampling and CFG Scale of 4.

Use common SDXL dimensions such as 1024x1024 and 896x1152.

@bobbelly404707 Which scheduler do you prefer?

@peenusbutter Those settings are bare minimum and are already on civitai, what I mean are technical specifications like optimal samplers and prompts to evoke different styles etc. No one should be post models so vaguely. But its all good, I figured out what is working nicely with this model.

@smokeymillz and like the creator, you couldn't be bothered to put down some technical details on what you got to work well?

@smokeymillz I leave it on Automatic which appears to pick Exponential. Karras, SGM Uniform and Simple also seem to work well.

@ShartHouse From the BigASP page "Finally, scoring. bigASP 🐍 v2, like v1 and inspired by the incredible work of PonyDiffusion, was trained with "score tags". This means it understands things like score_8_up, which specify the quality of the image you want generated. All images in bigASP's dataset were scored from 0 to 9, with 0 completely excluded from training, and 9 being the best of the best. So when you write something like score_7_up at the beginning of your prompt, you're telling bigASP "I want an image that's at least a quality of 7."

@blaqua thank you for explaining this shit. I have never figured this out and been wondering but didn't want to look like an idiot asking

Search for BigASP2Lustify that uses latest versions + better balance IMO

@ronikush That one hasn't been updated since October of last year, it's even older than this is

check for biglust 1.7 which is a merge of:

BigASP 2 (60% UNET / 100% CLIP)

Lustify OLT with texture fix (40% UNET / 0% CLIP)

Monkey see, monkey do. There's so much vomit in some of these prompts that it does more harm than good. Pony tags, chatgpt essays, etc.

Maybe they are using the model as a refiner

It's a merge of bigASP which was trained using the same score tags as Pony.

Many people generate first pass in Pony for its creative and expressive scenes. Then do a second pass like Lustify or BigLust to bring out the realism.

@damiang This. it isn't completely monkey see monkey do. the "KitBash" or what have you gets very good results. I often mess with models here then throw the generated images onto my local machine and img2img with flux or another model without onsite generation. pony is also so much less buzz than flux so I think it is smart to start there.

This is a merge of bigASP and LUSTIFY.And if you read the Features of bigASP,

this model understands quality scores to help improve generations, e.g. add score_7_up, to the start of your prompt to guide the quality of generations. More details below.

I used the civitai trainer and it works well with big lust, training locally with FluxGym also gives good results.

i use khoya_ss in win11 wsl2 with a 4090. start with a prodigy template in the lora section and a network dim of 32 that will produce 230mb loras. you can reduce the network dim and get smaller sizes you have to experiment if it still does what you want. https://github.com/bmaltais/kohya_ss

I have the same problem!, There is a tutorial or something??

this may sound dumb but how do i use this in stability mastrix?

Can you help me? I’m not sure what I’m doing wrong. What settings did you tweak and prompt tweaks?

@Chasris what you mean? you can find a great workflow from just save an image and then dropping it into comfy ui.

can you share your other best findings please, i am lost on too many models

@pink0909 Sure, but are you looking for realistic? I haven't been doing much realistic lately, more so a mix of realism. I like the illustrijEVO v2 and v3, illustrij v18, mergeijPONYIL v3, oneObession v17, vanimix v15, waiNSFWIllustrious v14, and molKeunMix deepblue V2EPS. If you want more realistic I have messed around with cyberrealistic models, cyberillustrious, realismByStableYogi and the illustrious version as well. Hope that helps!

in the description here click on bigasp, read the description there. use bigasp's recommended settings or just copy an image here from civitai directly into your comfyui workflow

Do a screen grab/screenshot and drop it in ChatGPT! You really don't even have to say/type anything! It'll recognize what it is then tell you what's wrong and how to fix it.

@boates1970 

@boates1970  Getting the same error. I literally just dropped the model in said go, and it said fuck you.

Possibly the download failed and left part of the file there. Try deleting it and re-downloading the model.

I guess those makers are using comfyUI ,and the prompts are for the first step image for later video generation.

It works, but only if the author has saved the metadata with the image or video. I find about half of what I download from here has metadata.

Only PNG Images can contain extra MetaData in the form of JSON Code because only PNG files have support for extra data fields for storage, so if you drag a JPG, nothing will ever happen. Any PNG on CivitAI will contain SOME extra data, usually some settings like Steps, CFG, Sampler Name etc, but only SOME OTHERS will contain a full JSON Workflow that ComfyUI can pull from. There is an AddOn for Firefox called "Image Metadata Viewer" - I assume other browsers might have something similar - that you can use to right click an any image in your browser window to see the MetaData. After you do that a few times you will learn exactly what to look for, which is a bit more convenient then having ComfyUI always open to drag test images into.

This is a checkpoint not a LORA

.. get a life kid..

I can show you how to train lora. For 20$

If you want hahahahahaha

click on the image above of the brunette girl spreading her ass to open it up fully. right click save as. It should try to save it as a .png (not a .webp), drag and drop into comfy.

@jdm55555 thanks